Computer Science | Grinnell College

On Thursday, November 20, Toby Baratta 2017, Bo Wang 2016, and Kitt Nika 2016 will present their summer research on the automatic detection of place names on historical maps:

This past summer we did research on toponym detection and recognition on historical maps. Overall, our research goal was making historical maps more search-friendly and making information in the maps more accessible. Kitt Nika and Shen Zhang worked on detecting text strings on map images with maximally stable extremal regions (MSER). With this, they implemented a binarization method for future research in text detection. Kitt will discuss MSERs and the methods involving them in regards to text detection. Toby Baratta and Bo Wang worked on linking geographic datasets and recognizing toponyms from the detected text strings. With the alignment between historical maps and real-life geography, we used a Bayesian model to calculate probabilities of possible toponyms. Toby and Bo will discuss their work on increasing the range of our recognition system by adding area features such as lakes.

At 4:15 p.m., refreshments will be served in the Computer Science Commons. The talk, “Historical map processing: text detectors, database linking, and region models,” will begin at 4:30 p.m. in Noyce 3821. Everyone is welcome to attend!

On Thursday, October 6, Zach Butler '13 and Dugan Knoll '12, will present a talk in the "Thursday Extra" series on their summer research:

Scientists have been working in the field of text recognition, the science of automatically reading text, for over 200 years. While the problem of reading whole documents (commonly called OCR, or optical character recognition) is more or less solved, the problem of reading text from arbitrary real-world scenes (Scene Text Recognition, or STR) still presents researchers with many challenges. Yet humans have been able to read such text ever since we created language. Many have created a robust recognition programs, but some still suffer from not knowing where the text baseline is—that is, where the non-descending characters of a line of text end. In this talk, we will discuss what makes reading scene text so difficult, how we made a baseline detection algorithm to improve the results of scene text recognition systems, and how we used the scientific method to make our system as robust as possible in ten weeks.

Refreshments will be served at 4:15 p.m. in the Computer Science Commons (Noyce 3817). The talk, "A robust system for discovering text baselines in scene text images," will follow at 4:30 p.m. in Noyce 3821. Everyone is welcome to attend.

On Thursday, April 7, Ravi Chande 2011 and Dylan Gumm 2011 will give a talk in the “Thursday Extras” series:

Maps -- particularly old ones -- have proven a significant challenge to traditional optical character recognition systems. Intersections of text with other cartographic symbols, irregular word orientations, and unusual fonts all serve as sources of error. Previous attempts to improve OCR on maps have used pre-processing steps on the input image or various post-processing techniques on the strings output by OCR. Here, we present a system that attempts to combine an OCR system with the known geography of the map to improve text recognition.

Refreshments will be served at 4:15 p.m. in the Computer Science Commons (Noyce 3817). The talk, “Text recognition on historical maps,” will follow at 4:30 p.m. in Noyce 3821. Everyone is welcome to attend!

On Thursday, February 10, Jerod Weinman will discuss some aspects of his recent work on text recognition:

Is your smart phone smarter than a fifth grader? Not yet. Accurately translating a photograph of text into an intrinsically textual representation has been confounding computational scientists for over a century. Humans (even fifth graders) still outperform computers at reading. In this talk, I review why the problem is difficult and present a model for robustly recognizing small amounts of text in images.

Refreshments will be served at 4:15 p.m. in the Computer Science Commons (Noyce 3817). Mr. Weinman's talk, “Robust text recognition,” will follow at 4:30 p.m. in Noyce 3821. Everyone is welcome to attend!